Cluster target classification and identification method and equipment based on multi-dimensional feature extraction

Abstract

The invention discloses a cluster target classification and identification method and device based on multi-dimensional feature extraction, and relates to the technical field of radar target identification. The method comprises the following steps: receiving an echo signal of a cluster target, and performing pulse compression on the echo signal to obtain data after pulse compression; based on a predetermined distance unit, performing bispectrum analysis on the data after pulse compression to obtain bispectrum estimation; performing feature extraction on the bispectrum estimation to construct a three-dimensional feature vector; and inputting the three-dimensional feature vector into a classification and recognition model so as to classify and recognize the cluster target. According to the invention, the identification capability of cluster targets and the universality of radar target group identification are improved.

Description

Technical Field

[0001] This invention relates to the field of radar target recognition technology, specifically to a method and device for classifying and recognizing clustered targets using multi-dimensional feature extraction. Background Technology

[0002] Cluster warfare, with its disruptive and innovative combat forms, rapid development, and enormous combat potential, has become a significant threat on the future battlefield, posing a challenge to traditional radar identification technologies. The diverse types of cluster targets, their dense distribution, complex topological distribution, and the mutual coupling of electromagnetic scattering all challenge the radar's anti-jamming capabilities, identification accuracy, and processing power. Therefore, radars must possess more intelligent and accurate identification capabilities for complex types of cluster targets.

[0003] Existing radar target classification and recognition technologies are focused on single targets. Furthermore, low-resolution radar is limited by extracting one-dimensional range profile features, while low-repetition-rate radar is limited by extracting micro-Doppler features. Summary of the Invention

[0004] This invention addresses the problem that existing radar target classification and recognition technologies are mainly focused on single targets or the difficulty of identifying complex groups of targets. It provides a multi-dimensional feature extraction-based method and device for classifying and recognizing clustered targets, which improves the recognition capability of clustered targets and the universality of radar target group recognition.

[0005] The present invention is achieved through the following technical solution.

[0006] Firstly, a cluster target classification and recognition method based on multi-dimensional feature extraction is provided, the method comprising:

[0007] Receive the echo signal from the cluster target and perform pulse compression on the echo signal to obtain pulse-compressed data;

[0008] Based on a predetermined distance unit, bispectral analysis is performed on the pulse-compressed data to obtain bispectral estimation;

[0009] Feature extraction is performed on the bispectral estimation to construct a three-dimensional feature vector;

[0010] The three-dimensional feature vector is input into the classification and recognition model to classify and recognize the cluster targets.

[0011] In some embodiments, receiving echo signals from cluster targets and performing pulse compression on the echo signals includes:

[0012] The received echo signal is subjected to a Fourier transform in the fast time domain to obtain the frequency domain echo signal.

[0013] Substituting the frequency domain echo signal into the distance compression formula, the pulse-compressed data is calculated.

[0014] In some embodiments, bispectral analysis is performed on the pulse-compressed data based on a predetermined distance unit to obtain a bispectral estimate, including:

[0015] For the pulse-compressed data, a predetermined distance unit is selected;

[0016] The pulse-compressed data within each predetermined distance unit is sampled in both the fast and slow time dimensions.

[0017] The sampled data is segmented, and a discrete Fourier transform is performed on each segment.

[0018] Calculate the triple correlation results of the discrete Fourier transform results of each segment of data;

[0019] The triple correlation results of the discrete Fourier transform of each segment of data are averaged to obtain the bispectral estimate of each segment of data.

[0020] In some embodiments, feature extraction is performed on the bispectral estimation to construct a three-dimensional feature vector, including:

[0021] Based on the bispectral estimation, the bispectral entropy, first moment, and second moment are calculated respectively.

[0022] The bispectral entropy, the first moment, and the second moment are combined to construct the three-dimensional feature vector.

[0023] In some embodiments, the method further includes: inputting the three-dimensional feature vector into a classification and recognition model to train the classification and recognition model before classifying and recognizing the cluster targets, wherein training the classification and recognition model includes:

[0024] Based on multiple predetermined distance units, bispectral analysis is performed on the pulse-compressed data to obtain bispectral estimation;

[0025] Feature extraction is performed on the bispectral estimation of the plurality of predetermined distance units to construct a three-dimensional feature vector for each predetermined distance unit;

[0026] A multi-sample feature set is constructed based on the three-dimensional feature vectors for the multiple predetermined distance units;

[0027] The classification and recognition model is trained based on the multi-sample feature set, wherein the classification and recognition model is a classification and recognition model based on a feedforward neural network.

[0028] Secondly, a cluster target classification and recognition device with multi-dimensional feature extraction is provided, the device comprising:

[0029] The pulse compression module is used to: receive the echo signal from the cluster target and perform pulse compression on the echo signal to obtain pulse-compressed data;

[0030] The bispectral estimation module is used to: perform bispectral analysis on the pulse-compressed data based on a predetermined distance unit to obtain a bispectral estimate;

[0031] A three-dimensional feature vector construction module is used to: extract features from the bispectral estimation to construct a three-dimensional feature vector;

[0032] The classification and recognition module is used to input the three-dimensional feature vector into the classification and recognition model to classify and recognize the cluster targets.

[0033] In some embodiments, the pulse compression module includes:

[0034] The frequency domain echo signal acquisition unit is used to: perform Fourier transform on the received echo signal in the fast time domain to obtain the frequency domain echo signal;

[0035] The pulse compression data calculation unit is used to: substitute the frequency domain echo signal into the distance compression formula to calculate the pulse-compressed data.

[0036] In some embodiments, the bispectral estimation module includes:

[0037] The distance unit determining unit is used to: select a predetermined distance unit for the pulse-compressed data;

[0038] The data sampling unit is used to sample the pulse-compressed data within each predetermined distance unit in both the fast and slow time dimensions.

[0039] The data segmentation unit is used to: segment the sampled data and perform discrete Fourier transform on each segment;

[0040] The triple correlation calculation unit is used to calculate the triple correlation results of the discrete Fourier transform results of each segment of data.

[0041] The bispectral estimation calculation unit is used to: average the triple correlation results of the discrete Fourier transform results of each segment of data to obtain the bispectral estimate of each segment of data.

[0042] In some embodiments, the three-dimensional feature vector construction module includes:

[0043] The three-dimensional feature calculation unit is used to: calculate the bispectral entropy, the first moment, and the second moment based on the bispectral estimation.

[0044] A three-dimensional feature vector combination unit is used to combine the bispectral entropy, the first moment, and the second moment to construct the three-dimensional feature vector.

[0045] Thirdly, a cluster target classification and recognition device with multi-dimensional feature extraction is provided, the device comprising:

[0046] At least one processor;

[0047] At least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor, the instructions implementing the method described in any of the above when executed by the at least one processor.

[0048] Compared with existing technologies, this invention has the following advantages and beneficial effects: First, pulse compression is performed on the radar echo of the cluster target. Then, the range cell of interest for the cluster target is selected for bispectral transformation. Based on the bispectral transformation result, bispectral entropy, first moment, and second moment are extracted to construct a multidimensional feature vector. Finally, a BP neural network model is used to classify the multidimensional feature vector of the cluster target and output the cluster target identification result. This invention can extract the effective features of the cluster target, achieve accurate classification and identification of the cluster target, improve the identification capability of the cluster target, and enhance the universality of radar target group identification. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart of a cluster target classification and recognition method based on multidimensional feature extraction according to an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of the bispectral analysis results of a group of targets according to an embodiment of the present invention.

[0052] Figure 3 This is a schematic diagram of a multi-sample feature set according to an embodiment of the present invention.

[0053] Figure 4 This is a schematic diagram of the group target classification and identification results according to an embodiment of the present invention.

[0054] Figure 5 This is a structural block diagram of a cluster target classification and recognition device based on multidimensional feature extraction according to an embodiment of the present invention.

[0055] Figure 6 This is a schematic diagram of a cluster target classification and recognition device based on multidimensional feature extraction according to an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0057] This invention proposes a classification and identification scheme for cluster targets to solve the problem of difficult identification of complex types of cluster targets. It involves: extracting the bispectral entropy, first moment, and second moment features of cluster targets. These parameters are independent of radar system parameters and have universal adaptability. At the same time, classification and identification are performed based on the extracted features, and the recognition rate can meet the requirements of general radar target identification.

[0058] To facilitate the description of the present invention, the following parameters are first defined: distance-to-time. Direction and slow time electromagnetic wave propagation speed linear frequency modulation slope carrier frequency Transmit signal bandwidth carrier wavelength Pulse repetition period Fast sampling interval .

[0059] On the one hand, the present invention provides a cluster target classification and recognition method based on multidimensional feature extraction. Figure 1 This is a flowchart illustrating a cluster target classification and recognition method based on multi-dimensional feature extraction according to an embodiment of the present invention. (Reference) Figure 1 The multidimensional feature extraction cluster target classification and recognition method includes S10 to S40.

[0060] The following describes S10 to S40 in detail with reference to the accompanying drawings.

[0061] In S10, the echo signal from the cluster target is received, and the echo signal is pulse compressed to obtain pulse-compressed data.

[0062] For example, receiving the echo signal from the cluster target and performing pulse compression on the echo signal includes: performing a Fourier transform on the received echo signal in the fast time domain to obtain a frequency domain echo signal; then, substituting the frequency domain echo signal into the range compression formula to calculate the pulse-compressed data.

[0063] Assume the echo signal of the cluster target is:

[0064] ; (1)

[0065] Among them, the above four items are: distance-oriented envelope directional envelope Transmitted signal phase and Doppler modulation term. For complex numbers, For the first Historical slant range of each target The target total number.

[0066] The distance compression process is represented as:

[0067] (2)

[0068] in, The response function of the reference function. For the Fourier transform of the echo in the fast time domain, Given a fast time frequency. Substituting equation (1) into equation (2) after performing a Fourier transform in the fast time domain, we obtain the distance compression result as follows:

[0069] (3)

[0070] in, This is the distance pulse compression response function.

[0071] In S20, bispectral analysis is performed on the pulse-compressed data based on a predetermined distance cell to obtain a bispectral estimate.

[0072] For example, bispectral analysis is performed on pulse-compressed data based on predetermined distance units to obtain bispectral estimates, including: selecting predetermined distance units for the pulse-compressed data; sampling the pulse-compressed data within each predetermined distance unit in both the fast and slow time dimensions; segmenting the sampled data and performing discrete Fourier transforms on each segment; calculating the triple correlation results of the discrete Fourier transform results for each segment; and averaging the triple correlation results of the discrete Fourier transform results for each segment to obtain bispectral estimates for each segment.

[0073] Specifically, for the pulse-compressed data, a distance-of-interest cell is selected. , recorded as Then, Discretization (sampling) results in , , and These represent the number of sampling points in the fast time dimension and the slow time dimension, respectively. Divided into, for example, part, Each paragraph contains There are 100 samples, that is, 100 samples For the first part Perform a discrete Fourier transform:

[0074] (4)

[0075] in, For frequency, , length is ; The slow sampling rate is the pulse repetition frequency. .

[0076] Calculate the triple correlation of equation (4) to obtain the bispectral estimate. ,Will Substituting into equation (5), we get Bispectral estimation :

[0077] (5)

[0078] in, .

[0079] In S30, feature extraction is performed on the bispectral estimation to construct a three-dimensional feature vector.

[0080] For example, feature extraction from bispectral estimation to construct a three-dimensional feature vector includes: calculating the bispectral entropy, first moment, and second moment based on the bispectral estimation; and then combining the bispectral entropy, first moment, and second moment to construct a three-dimensional feature vector.

[0081] Specifically, the bispectral entropy is calculated using the following formula. :

[0082] (6)

[0083] (7)

[0084] The first moment is calculated using the following formula. ;

[0085] (8)

[0086] (9)

[0087] The second moment is calculated using the following formula. :

[0088] (10)

[0089] Will , as well as Combining to construct three-dimensional feature vectors .

[0090] In S40, three-dimensional feature vectors are input into the classification and recognition model to classify and recognize cluster targets.

[0091] In some embodiments, the multidimensional feature extraction cluster target classification and recognition method further includes: inputting a three-dimensional feature vector into a classification and recognition model to train the classification and recognition model before classifying and recognizing the cluster target.

[0092] For example, training the classification and recognition model includes: performing bispectral analysis on pulse-compressed data based on multiple predetermined distance units to obtain bispectral estimates; extracting features from the bispectral estimates of the multiple predetermined distance units to construct a three-dimensional feature vector for each predetermined distance unit; constructing a multi-sample feature set based on the three-dimensional feature vectors of the multiple predetermined distance units; and training the classification and recognition model based on the multi-sample feature set. The classification and recognition model is a classification and recognition model based on a feedforward (BP) neural network. Specifically, the multi-sample feature set can be... Input the backpropagation (BP) neural network, set the effective sample size, test sample size, and number of hidden layer neurons, and then train it. (Multi-sample feature set) By selecting Q distance units and performing operations S20, S30, and S40, their corresponding feature vectors can be obtained, thereby constructing a multi-sample feature set. .

[0093] The method of the present invention will be verified and illustrated through specific application examples below.

[0094] Based on the MATLAB platform, the target vehicle group was obtained according to the simulation parameters shown in Table 1. Bird flock target and drone swarm targets The echo data.

[0095] Table 1

[0096]

[0097] Step 1: Pulse Compression

[0098] Each , as well as Substituting into equation (2) and performing pulse compression processing, we obtain equation (3). , as well as .

[0099] Step 2: Bispectral Analysis

[0100] Data after pulse compression , Select the first distance cell and perform bispectral analysis, where... The value was set to 8. The bispectral data for vehicle flocks, bird flocks, and drone flocks were obtained as follows: Figure 2 As shown in (a), (b), and (c).

[0101] Step 3: Feature Extraction

[0102] Features are extracted from the output of step two according to equations (6) to (10). , as well as Combining to construct three-dimensional feature vectors .

[0103] For each of the vehicle flock, bird flock, and drone swarm, 500 distance units were selected. Following steps two, three, and four, their corresponding feature vectors were obtained to construct a multi-sample feature set. Multi-sample feature sets, such as Figure 3 As shown in (a), (b), and (c).

[0104] Step 4: Classification and Recognition

[0105] The multi-sample feature set in step three A backpropagation (BP) neural network is input, with the effective sample size set to 25%, the training sample size to 60%, the test sample size to 15%, and the number of hidden layer neurons to 10. After setting these parameters, training is performed to obtain a classification and recognition model. Inputting the three-dimensional feature vector into the classification and recognition model allows for the classification and recognition of clustered targets. Classification and recognition results for single features and multi-dimensional features are shown below. Figure 4As shown in the table, (a) represents the classification and recognition results based on "entropy," (b) represents the classification and recognition results based on "first moment," (c) represents the classification and recognition results based on "second moment," and (d) represents the classification and recognition results based on three-dimensional features. The recognition rates are summarized in Table 2. It can be seen that the recognition rate based on bispectral entropy features is 62.1%, the recognition rate based on bispectral first moment is 54.7%, the recognition rate based on bispectral second moment is 54.3%, and the recognition rate based on all three features is 82.7%. Therefore, combining the three types of features for cluster target classification and recognition significantly improves the recognition rate.

[0106] Table 2

[0107]

[0108] In summary, by extracting bispectral entropy, first moment, and second moment from cluster targets and constructing feature vectors, a BP neural network can be used to effectively classify and identify cluster targets. Simulation results verify the effectiveness of the algorithm.

[0109] In this invention, the cluster target classification and recognition method based on multi-dimensional feature extraction mainly includes the following steps: echo preprocessing, multi-dimensional feature extraction based on bispectral transform, and classification and recognition based on a BP neural network. Echo preprocessing includes pulse compression of the baseband echo data acquired by radar to obtain the input for bispectral analysis. Multi-dimensional feature extraction based on bispectral transform includes selecting the range cell of interest, performing bispectral transform in the slow-time dimension to obtain the bispectral statistical characteristics in the slow-time dimension, and extracting entropy, first moment, and second moment features based on the bispectral statistical results. Bispectral transform can effectively suppress the influence and obscuring of external Gaussian noise on the signal itself; extracting bispectral entropy features can effectively reflect the magnitude of signal energy divergence along the frequency dimension; extracting first and second moment features can effectively reflect the phase features in the image after bispectral transform. Bispectral entropy, first moment, and second moment can all reflect the characteristics of cluster targets. Finally, using a classification and recognition method based on a BP (feedforward) neural network, the extracted multi-dimensional features of cluster targets are input to complete the effective classification and recognition of different target groups.

[0110] On the other hand, the present invention provides a cluster target classification and recognition device with multi-dimensional feature extraction. Figure 5 This is a structural block diagram of a cluster target classification and recognition device based on multi-dimensional feature extraction according to an embodiment of the present invention. (Reference) Figure 5 The multidimensional feature extraction cluster target classification and recognition device includes: a pulse compression module, a bispectral estimation module, a three-dimensional feature vector construction module, and a classification and recognition module.

[0111] The pulse compression module is used to receive the echo signal from the cluster target and perform pulse compression on the echo signal to obtain pulse-compressed data.

[0112] The bispectral estimation module is used to perform bispectral analysis on pulse-compressed data based on a predetermined distance cell to obtain a bispectral estimate.

[0113] The 3D feature vector construction module is used to extract features from bispectral estimation to construct 3D feature vectors.

[0114] The classification and recognition module is used to input three-dimensional feature vectors into the classification and recognition model to classify and recognize cluster targets.

[0115] In some embodiments, the pulse compression module includes: a frequency domain echo signal acquisition unit, configured to: perform a Fourier transform on the received echo signal in the fast time domain to obtain a frequency domain echo signal; and a pulse compression data calculation unit, configured to: substitute the frequency domain echo signal into the distance compression formula to calculate the pulse-compressed data.

[0116] In some embodiments, the bispectral estimation module includes: a distance unit determination unit, configured to: select a predetermined distance unit for the pulse-compressed data; a data sampling unit, configured to: sample the pulse-compressed data within each predetermined distance unit in both the fast and slow time dimensions; a data segmentation unit, configured to: segment the sampled data and perform a discrete Fourier transform on each segment; a triple correlation calculation unit, configured to: calculate the triple correlation result of the discrete Fourier transform results of each segment; and a bispectral estimation calculation unit, configured to: average the triple correlation results of the discrete Fourier transform results of each segment to obtain a bispectral estimate of each segment.

[0117] In some embodiments, the three-dimensional feature vector construction module includes: a three-dimensional feature calculation unit, used to: calculate the bispectral entropy, the first moment, and the second moment respectively based on bispectral estimation; and a three-dimensional feature vector combination unit, used to: combine the bispectral entropy, the first moment, and the second moment to construct a three-dimensional feature vector.

[0118] In some embodiments, the multidimensional feature extraction cluster target classification and recognition device further includes: a classification and recognition model training module, configured to: perform bispectral analysis on pulse-compressed data based on multiple predetermined distance units to obtain bispectral estimates; extract features from the bispectral estimates of the multiple predetermined distance units to construct a three-dimensional feature vector for each predetermined distance unit; construct a multi-sample feature set based on the three-dimensional feature vectors for the multiple predetermined distance units; and train the classification and recognition model based on the multi-sample feature set, wherein the classification and recognition model is a classification and recognition model based on a feedforward neural network.

[0119] For other implementation details of the multidimensional feature extraction cluster target classification and recognition device, please refer to the previous description of the multidimensional feature extraction cluster target classification and recognition method, which will not be repeated here.

[0120] In implementing the functions of the integrated modules described above in hardware, this embodiment of the invention provides a structure for a cluster target classification and recognition device involving multi-dimensional feature extraction as described in the above embodiments. Figure 6 This is a schematic diagram of a cluster target classification and recognition device based on multi-dimensional feature extraction according to an embodiment of the present invention. (Reference) Figure 6 The multidimensional feature extraction cluster target classification and recognition device includes: at least one processor; and at least one memory. The at least one memory is coupled to the at least one processor and stores instructions for execution by the at least one processor, which, when executed by the at least one processor, implement the method described above.

[0121] A processor can be a set of logic blocks, modules, and circuits that implement or execute the various exemplary logic blocks, modules, and circuits described in connection with embodiments of the present invention. The processor can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in connection with embodiments of the present invention. A processor can also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, etc.

[0122] The memory may be read-only memory (ROM) or other types of static storage devices that can store static information and instructions, random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.

[0123] In one implementation, the memory can exist independently of the processor. The memory can be connected to the processor via a bus and used to store instructions or program code. When the processor calls and executes the instructions or program code stored in the memory, it can implement the method provided in the embodiments of the present invention. In another implementation, the memory can also be integrated with the processor.

[0124] On the other hand, the present invention also provides a computer-readable storage medium (e.g., a non-transitory computer-readable storage medium) storing computer program instructions that, when executed on a computer, cause the computer to perform the method as described in any of the above embodiments.

[0125] Exemplary examples show that the aforementioned computer-readable storage media may include, but are not limited to: magnetic storage devices (e.g., hard disks, floppy disks, or magnetic tapes), optical discs (e.g., compact disks (CDs), digital versatile disks (DVDs), etc.), smart cards, and flash memory devices (e.g., erasable programmable read-only memory (EPROMs), cards, sticks, or key drives, etc.). The various computer-readable storage media described in this invention may represent one or more devices and / or other machine-readable storage media for storing information. The term "machine-readable storage medium" may include, but is not limited to, wireless channels and various other media capable of storing, containing, and / or carrying instructions and / or data.

[0126] This invention provides a computer program that, when run on a computer, causes the computer to perform the method of any of the above embodiments.

[0127] This invention provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the method of any of the above embodiments.

[0128] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A cluster target classification and recognition method with multi-dimensional feature extraction, characterized in that, The method includes: Receive the echo signal from the cluster target and perform pulse compression on the echo signal to obtain pulse-compressed data; Based on a predetermined distance unit, bispectral analysis is performed on the pulse-compressed data to obtain bispectral estimation; Feature extraction is performed on the bispectral estimation to construct a three-dimensional feature vector; The three-dimensional feature vector is input into the classification and recognition model to classify and recognize the cluster targets.

2. The method according to claim 1, characterized in that, Receiving echo signals from cluster targets and performing pulse compression on the echo signals includes: The received echo signal is subjected to a Fourier transform in the fast time domain to obtain the frequency domain echo signal. Substituting the frequency domain echo signal into the distance compression formula, the pulse-compressed data is calculated.

3. The method according to claim 1 or 2, characterized in that, Based on a predetermined distance unit, bispectral analysis is performed on the pulse-compressed data to obtain bispectral estimation, including: For the pulse-compressed data, a predetermined distance unit is selected; The pulse-compressed data within each predetermined distance unit is sampled in both the fast and slow time dimensions. The sampled data is segmented, and a discrete Fourier transform is performed on each segment. Calculate the triple correlation results of the discrete Fourier transform results of each segment of data; The triple correlation results of the discrete Fourier transform of each segment of data are averaged to obtain the bispectral estimate of each segment of data.

4. The method according to claim 1, characterized in that, Feature extraction is performed on the bispectral estimation to construct a three-dimensional feature vector, including: Based on the bispectral estimation, the bispectral entropy, first moment, and second moment are calculated respectively. The bispectral entropy, the first moment, and the second moment are combined to construct the three-dimensional feature vector.

5. The method according to claim 1, characterized in that, The method further includes: inputting the three-dimensional feature vector into a classification and recognition model to train the classification and recognition model before classifying and recognizing the cluster targets, wherein training the classification and recognition model includes: Based on multiple predetermined distance units, bispectral analysis is performed on the pulse-compressed data to obtain bispectral estimation; Feature extraction is performed on the bispectral estimation of the plurality of predetermined distance units to construct a three-dimensional feature vector for each predetermined distance unit; A multi-sample feature set is constructed based on the three-dimensional feature vectors for the multiple predetermined distance units; The classification and recognition model is trained based on the multi-sample feature set, wherein the classification and recognition model is a classification and recognition model based on a feedforward neural network.

6. A cluster target classification and recognition device based on multi-dimensional feature extraction, characterized in that, The device includes: The pulse compression module is used to: receive the echo signal from the cluster target and perform pulse compression on the echo signal to obtain pulse-compressed data; The bispectral estimation module is used to: perform bispectral analysis on the pulse-compressed data based on a predetermined distance unit to obtain a bispectral estimate; A three-dimensional feature vector construction module is used to: extract features from the bispectral estimation to construct a three-dimensional feature vector; The classification and recognition module is used to input the three-dimensional feature vector into the classification and recognition model to classify and recognize the cluster targets.

7. The device according to claim 6, characterized in that, The pulse compression module includes: The frequency domain echo signal acquisition unit is used to: perform Fourier transform on the received echo signal in the fast time domain to obtain the frequency domain echo signal; The pulse compression data calculation unit is used to: substitute the frequency domain echo signal into the distance compression formula to calculate the pulse-compressed data.

8. The device according to claim 6 or 7, characterized in that, The bispectral estimation module includes: The distance unit determining unit is used to: select a predetermined distance unit for the pulse-compressed data; The data sampling unit is used to sample the pulse-compressed data within each predetermined distance unit in both the fast and slow time dimensions. The data segmentation unit is used to: segment the sampled data and perform discrete Fourier transform on each segment; The triple correlation calculation unit is used to calculate the triple correlation results of the discrete Fourier transform results of each segment of data. The bispectral estimation calculation unit is used to: average the triple correlation results of the discrete Fourier transform results of each segment of data to obtain the bispectral estimate of each segment of data.

9. The device according to claim 6, characterized in that, The three-dimensional feature vector construction module includes: The three-dimensional feature calculation unit is used to: calculate the bispectral entropy, the first moment, and the second moment based on the bispectral estimation. A three-dimensional feature vector combination unit is used to combine the bispectral entropy, the first moment, and the second moment to construct the three-dimensional feature vector.

10. A cluster target classification and recognition device based on multi-dimensional feature extraction, characterized in that, The device includes: At least one processor; At least one memory coupled to the at least one processor and storing instructions for execution by the at least one processor, the instructions implementing the method of any one of claims 1 to 5 when executed by the at least one processor.

Images